How we hire engineers in the AI era: no algorithm puzzles, no tool bans, and one interview that reveals everything

We dropped the “reverse a linked list” test in 2024. It measured something we no longer needed, and missed everything we did. Since then we have hired dozens of engineers for projects in eight countries, and this is the method we settled on — it may not suit everyone, but it works.

The principle: measure what they will do at work, with the tools they will use at work

Our engineers work with Claude Code and Cursor daily. Banning them in an interview is like banning a calculator in an accountant’s interview. Tools are allowed — required, even — and the question changes: not “can you write this?” but “can you get to the right thing, verify it, and explain it?”.

The one interview (90 minutes)

1. A real (anonymised) client problem: “A salon wants WhatsApp appointment reminders with attendance confirmation; the current system is Laravel.” No more detail. The first 15 minutes we listen to the questions they ask — questions reveal more than answers.
2. Build with the agent (40 minutes): they open their preferred tool and work. We watch: how do they describe the task? Do they read what is produced or accept it? Do they notice the agent forgot the cancellation case? Do they test?
3. The defence (20 minutes): we ask “why this table?”, “what if WhatsApp sends the message twice?”, “how do you know it works a month later?”. Whoever cannot explain their code does not own it — even if they typed it themselves.
4. Writing (15 minutes): a paragraph for a non-technical client explaining what was built, what was not, and why. Arabic or English depending on the assumed client.

We do not hire the fastest coder. We hire whoever notices the mistake fastest.

Exactly what we look for

• Domain curiosity: they ask about WhatsApp’s 24-hour window, the cancellation policy, the branches. They think like a product owner.
• Healthy scepticism: they read the agent’s output as a reviewer, not a consumer.
• Trust boundaries: they ask where the phone number comes from and whether it is verified. Security as instinct, not a checklist.
• Clarity: they say “I don’t know, I’ll check” instead of confident guessing.

What we do not care about

The university. The number of languages on the CV. Memorised algorithmic complexity. Cloud certificates. All can be fine, but none predicts anything we see in the interview.

For applicants: how to prepare

1. Build a small, complete project with an agent tool, ship it, and be ready to defend every decision in it.
2. Pick a domain and read its regulations (fintech, health, real estate). You will understand our questions differently.
3. Practise explaining a technical idea to a non-technical person in five sentences.

And for anyone still looking for their path before applying, we built platforms like MyCareer and Career Scale for exactly that: knowing what fits you before trying to fit everything.